In many cases, measurements at an array of sensors are the linear combinations of source signals. Usually the original source signals are wanted and thus need to be extracted from the linear mixtures of the source signals. The above problem is designated as blind separation of sources or as waveform -preserving blind estimation of multiple independent sources because the problem is to estimate the multiple source signals from an observations without knowledge of the transmission channels and source signals.
Previous works are surveyed and their limitations are examined, finally a new learning algorithm based on the all order cumulant is proposed. To increase learning speed, we modified the adaptation rule. The simulation results are given to verify the validity and advantage of the new algorithm.